Thinking without AI
Legal AI relieves legal professionals of the very tasks through which their expertise develops. Therein lies a paradox: those who increasingly delegate intellectual work risk eroding the skills required to review the output. How can this newfound efficiency be harnessed without undermining one's own professional judgment and the training of the next generation of lawyers?
I am building a legal AI platform. And that is precisely why I am writing this text.
It concerns a well-known phenomenon with a new name: cognitive offloading, the transfer of mental work to external tools.
Calculators, calendars, and navigation systems have relieved our memory and attention for decades. Such tools are not a sign of a lack of competence. They create space for more demanding tasks.
Legal AI is also cognitive offloading. A lengthy judicial decision no longer needs to be structured manually. A first draft is generated in seconds. Those who use the time saved to conduct deeper analyses, develop counterarguments, and sharpen strategy are using AI to amplify their own competence. Conversely, those who merely adopt the output do not simply outsource work; they outsource thinking itself.
Things become critical when occasional assistance turns into a habit. Those who regularly work with AI drafts never start from a blank page again. This is efficient. However, it alters the cognitive process: developing becomes reviewing; formulating becomes editing; structuring becomes reacting.
These tasks also require legal competence. Yet they do not train the same skills. And skills that are no longer practiced atrophy. This is known as deskilling. Most at risk is that which only develops and endures through repeated application: the analysis of complex sets of facts, precise legal subsumption, and legal drafting.
The particular risk of legal AI does not lie in its errors. It lies in how persuasive its outputs are. A cleanly drafted text conveys competence. A cited authority fosters trust—even when it does not exist or does not stand for the proposition asserted. The more persuasive the output, the lower the willingness to independently verify the underlying authorities. The lawyer shifts from author to reviewer.
Yet review requires competence. Only those who can analyze independently will recognize whether AI has overlooked a decisive issue. Only those who can construct arguments themselves will see through reasoning that is merely superficially compelling. Therein lies the paradox: the more we rely on AI, the more critical the ability to think without AI becomes.
Experienced attorneys have developed this ability over years. They possess benchmarks and professional intuition. For junior associates, this foundation is still in development. Much of what is considered routine work today was traditionally legal training: those who had to summarize a judgment themselves learned to distinguish the material from the immaterial. Those who conducted legal research from scratch learned not only the results, but also research methodologies, dissenting views, and dead ends. If AI takes over these tasks entirely, it is not merely tedious work that disappears; learning opportunities disappear. Law firms must harness productivity gains without undermining the training of those who will ultimately bear responsibility for AI-generated work product.
This does not warrant a rejection of legal AI. Technology has always reshaped legal practice, and no one is seriously advocating a return to the typewriter. AI can even foster competence: as a sparring partner that provides counterarguments, reveals blind spots, and offers immediate feedback. The decisive factor is not whether AI is used; the decisive factor is how.
A mere "human in the loop" is not sufficient here. A person who signs off on an AI-generated draft at the final stage remains only formally within the process; they have shaped neither the legal issue, nor the analysis, nor the arguments. Substantive oversight means more: precisely framing the initial issue oneself, independently verifying authorities, developing counterarguments, determining strategy, and assuming full responsibility for the work product. This requires a comprehensive mastery of the case file that cannot be delegated to an AI.
Ultimately, deskilling is not only an individual risk, but an organizational issue. A law firm that exclusively rewards speed obscures whether the cognitive process behind the final work product is still mastered. What is needed is a culture where critical review is not regarded as a waste of time. For the review of an AI-generated output is not a mere formality; it is an independent legal accomplishment in its own right.
The deployment of legal AI cannot be stopped. The efficiencies are simply too substantial. The relevant question, therefore, is not whether legal professionals should use AI. It is: How do we use AI without losing the very skills we need to oversee it?
Drafted by
